Assignment and balance need to coordinate; assigner gives out up to 10 regions at a time unbalancing a balanced cluster making for churn, etc.

Assigner needs to take into account current region balance. I'm seeing on recovery from a crash on a big cluster that we'll give near ten regions to first regionserver that reports in. If that regionserver successfully opens the assigned ten, then that'll put it over the balancer slop differential and it'll then be asked to close its regions by the balancer. This adds to the regionserver crash churn. Dumb.